Chaos theory often points in the opposite direction. For example, consider weather simulations which become worse than careful ignorance after 5 days- slight variations in initial conditions (and multiplication roundoff errors in computation, and so on) grow out of control, and soon the system is less accurate than just saying “it rains 20% of the time in general; 6 days from now, there is a 20% chance of rain.”
that short-term projections can be terribly inaccurate, even while long-term forecasting can be extremely accurate?
It is often the case that long-run means are easier to predict than short-run means, in large part because the variability in long-run means is lower. This is especially the case for systems with negative feedback loops, where the system corrects deviations from normality, making normality especially likely.
It’s not clear to me that that does much for oil prices or social security deficits, since I don’t see either as being systems where the negative feedback is obviously stronger than the positive feedback.
Typically, short-term forecasting is stymied by noise rather than fundamental underlying uncertainty. For example, consider the wager between Simon and Ehrlich. They used a basket of commodities because they didn’t want short-term noise to upset the wager, but the main difference in the long-term predictions was the different underlying models.
In both the oil and Social Security examples, there are powerful long-term trends which mean we should have as much or more confidence in long-term projections than short-term ones: in oil, as a nonrenewable resource, the more efficient the market the closer it will conform to Hotelling’s rule, and in SS, it’s almost entirely driven by locked-in demographics or actuarial factors, and the uncertainty is in how and whether payouts will be modified or revenue increased.
(The latter might be what Taleb is getting at, but since he’s an arrogant blowhard who loves to oversimplify and believes he is right about everything, I am not inclined to be charitable and think he’s making a subtle claim about the different sources of variability and their foreseeability over the short and long run.)
Regardless, Taleb is making the argument: “if we cannot predict something in the short term, we cannot predict it in the long-term” which is not true of many things and may not even be true of his chosen examples.
Chaos theory often points in the opposite direction. For example, consider weather simulations which become worse than careful ignorance after 5 days- slight variations in initial conditions (and multiplication roundoff errors in computation, and so on) grow out of control, and soon the system is less accurate than just saying “it rains 20% of the time in general; 6 days from now, there is a 20% chance of rain.”
It is often the case that long-run means are easier to predict than short-run means, in large part because the variability in long-run means is lower. This is especially the case for systems with negative feedback loops, where the system corrects deviations from normality, making normality especially likely.
It’s not clear to me that that does much for oil prices or social security deficits, since I don’t see either as being systems where the negative feedback is obviously stronger than the positive feedback.
Typically, short-term forecasting is stymied by noise rather than fundamental underlying uncertainty. For example, consider the wager between Simon and Ehrlich. They used a basket of commodities because they didn’t want short-term noise to upset the wager, but the main difference in the long-term predictions was the different underlying models.
In both the oil and Social Security examples, there are powerful long-term trends which mean we should have as much or more confidence in long-term projections than short-term ones: in oil, as a nonrenewable resource, the more efficient the market the closer it will conform to Hotelling’s rule, and in SS, it’s almost entirely driven by locked-in demographics or actuarial factors, and the uncertainty is in how and whether payouts will be modified or revenue increased.
(The latter might be what Taleb is getting at, but since he’s an arrogant blowhard who loves to oversimplify and believes he is right about everything, I am not inclined to be charitable and think he’s making a subtle claim about the different sources of variability and their foreseeability over the short and long run.)
Regardless, Taleb is making the argument: “if we cannot predict something in the short term, we cannot predict it in the long-term” which is not true of many things and may not even be true of his chosen examples.